In previous Units we have looked at the materials used in thermal management, and given some examples of the thermal parameters of representative material. But we have said little about how these properties might be measured, and we have also taken for granted that parameters such as temperature and airflow can be measured easily and accurately. But in real life we need information on the properties of materials, and we need to be able to compare the results of simulation with what is achieved by the real system. So this Unit gives a brief overview of some of the measurement challenges related to thermal management.
In the short sections that follow, we examine some of the ways in which the temperature within an environment can be measured. Or it might be truer to say approximately measured, because the methods have inbuilt inaccuracies. Some of these inaccuracies are the result of offsets and non-linearities in the measurement device; others reflect the lack of sensitivity of the measuring system; yet other techniques give results that are actually misleading, because the act of making the measurement has actually changed the temperature within the system. As an example of this last effect, consider the challenge of measuring a small heat source with a sensor of high thermal mass. So, as you review each technique, try to estimate its likely accuracy and repeatability, as well as the impact of the measurement on the system under investigation.
As those of you who have carried out thermal profiling during reflow soldering will know, making even representative measurements by contact methods such as thermocouples is problematic. Apart from the inaccuracy of the thermocouple itself, it needs to be attached to the area of interest, and this inevitably changes the thermal situation. In reflow, with heat applied from the outside, we know that the thermal mass of the thermocouple is an additional load on the heat available, and this is exacerbated by the thermal mass of the resin, solder or tape used to attach the thermocouple. In the thermal measurement situation, the situation is even worse, because the presence of a thermocouple and its associated restraint may also have an impact on the local airflow. Nevertheless, appropriate use of thermocouples can give useful information on heat rise within an equipment, particularly if “informed common sense” is used to select the positions to be monitored, choosing some that are representative of cooler areas of the board as well as the most likely hot spots.
Robert Moffat Notes on using thermocouples.
We indicated in Unit 3 that materials have a temperature coefficient of resistance, and this can be used for temperature measurement. Resistance temperature detectors (RTDs) are wire-wound or thin-film devices capable of giving very precise and stable readings, but usually as encapsulated probes, so with higher thermal mass than thermocouples. As with thermocouples, the output from an RTD is low and needs amplification and conversion to give a temperature reading.
Thermistors are resistors with a much greater sensitivity to changes in temperature that are available with both negative and positive temperature coefficients. The latter, ‘PTC thermistors’, have almost a switching characteristic from low to high resistance at their critical temperature and are often used for safety cut-out applications; NTC types are better suited to temperature measurement or control. Typically based around doped ceramic/ferrite materials, thermistors are available in many different sizes and formats, some quite small, and are well-suited to permanent integration within a printed circuit assembly.
At temperatures.com, a repository of information about temperature measurements and devices, read the information on Resistance Temperature Detectors (RTDs) and Thermistor Temperature Sensors. At eFunda, read more about the Resistance-Temperature Relationship of RTDs. Both these web sites are well worth book-marking for additional information on measurement issues.
Liquid crystal materials will be familiar from their use in thermal paints, a “quick and dirty” contact method that is limited to a once-off basis. Available for a number of temperature ranges, these turn colour once they have reached a critical temperature. A possible application is to verify that nowhere on the board has exceeded a critical temperature. However, as with thermocouples and RTDs, where only a small number of temperatures is typically taken, thermal paints are rarely used on more than selected areas.
More information on thermal paints and indicators at the Thermax web site.
However, thermochromic liquid crystals can be used to give a graphic view of the thermal distribution over an area, and have been used both at the level of the silicon die and the PCB. This technique is particularly useful for hot spot detection, which is becoming important with integrated circuits; liquid crystal systems can be used at high resolution to investigate the heat distribution on a wafer held in a thermal test-head chuck.
Read Azar and Farina Measuring chip temperatures with thermochromic liquid crystals for a description of how the technology works, and then read Dino Farino’s Making surface temperature measurements using liquid crystal thermography for an insight into whether to select a narrow-band or wide-band formulation.
There are some pretty pictures, and more detailed information in Stasiek and Kowalewski Thermochromic liquid crystals applied for heat transfer research (PDF file, 829KB).
The review in Kaveh Azar’s Introduction to Liquid Crystal Thermography has some useful slides comparing different techniques for temperature measurement (PDF file, 224KB).
So far, all the methods we have looked at require the measurement device to be in contact with whatever we are examining. But we know from earlier studies that every body emits radiation (infrared or visible) and that the amount and frequency distribution of the radiation are determined by the body’s temperature. A pyrometer, or radiation thermometer, is a non-contact instrument that detects the radiation emitted from an object and uses this information to calculate its surface temperature.
Depending on the application and temperature range, a small number of specific wavelengths may be targeted (‘narrow-band’), or a wide range of wavelengths measured (‘broad-band’). Radiation thermometers have been constructed from several different pyroelectric materials, including lithium tantalate and polyvinylidene fluoride. One recent pyrometer uses silicon micro-machining to produce a thermopile of stacked silicon-aluminium elements to give improved sensitivity and accuracy within a small size.
A sensor of heat flow rather than directly of temperature, a thermopile has to be calibrated in order to produce temperature readings. It also needs compensation for ambient temperature and the emissivity of the surface, and it is quite common practice to coat part of the area being measured with a matt finish in order to approximate to a black body.
For a comprehensive review of pyrometry, read Jürgen Schilz Thermoelectric infrared sensors (thermopiles) for remote temperature measurements; pyrometry (PDF file, 378KB).
The pyrometer gives an accurate reading, but one that is averaged over a portion of the surface. For a view with more detail and visible thermal gradation, an infrared scanner can provide a fast and accurate picture of a complete board or assembly. IR thermography is made possible by a focal plane array, an assembly of closely-spaced infrared sensors, often held at very low temperature to increase its efficiency. A typical equipment might have a matrix of 320×240 pixels, with a temperature resolution of ±1°C around room temperature and ±2°C at higher temperatures.
Infrared scanner software can calibrate an object for emissivity variations based on taking an infrared picture of the unpowered product. For more insight into this issue read Jukka Rantala Emissivity in Practical Temperature Measurements.
But even infrared imaging, although a non-contact method that can give a true reflection of the temperature of the surface, is not able to measure what is probably the most critical temperature of all, which is the temperature of the semiconductor junction. Fortunately, the base-emitter characteristic of a transistor is temperature-dependent. Provided that the device has been characterised, measurement of this parameter can be converted to give an accurate reading of the junction temperature. In some cases, a separate discrete temperature-sensing device may be fitted within a circuit, but it is more typical for a test cell to be contained within the integrated circuit. Not only can this method be used for temperature measurement, but it can form the basis of an over-temperature protection circuit, where feedback from the sensing element is used either to shut down the component or to reduce the clock speed and hence the dissipation.
But, if your computer indicates a particular core temperature, don’t always assume that this is necessarily based on the device temperature; often a cheaper alternative is to use a standard microprocessor and place under it a heat-sensing element such as a thermistor. In this case, there may be some discrepancy between the temperature registered on the thermistor and the actual junction temperatures within the microprocessor.
Thermocouples can easily give information on heating and cooling rates, as well as on the peak steady state temperature, so that the thermal capacity of an assembly can be estimated. At the chip level, thermal transient testing uses a chip’s transient thermal behaviour to measure the quality of individual bonds between sections, and relies on the fact that the thermal time constants of the components that make up a semiconductor assembly – chip, substrate and package – typically differ from one another by at least an order of magnitude.
When power is first applied to the device, heat is generated at the junction and travels through the silicon, raising its temperature. However, when the heat flow reaches the next level of ‘heat sink’ in the package, further temperature rise in the silicon will be delayed until the heat sink is ‘saturated’. A plot of temperature against time (Figure 1) thus allows the thermal parameters of the different elements within the package to be shown separately.
If the integrity of the die mount has been affected by voids, the thermal resistance will be higher, the time required for the heat to reach the package heat sink will increase, and thus the final temperature of the silicon will be higher. The steps of the procedure are:
Figure 2 shows the result of testing two parts, one with known good die mount integrity, and one with suspected die-heat sink delamination following thermal shock testing. Both curves show a change in slope in ΔVbe with time, but the device with suspect die mount integrity exhibits a much higher ΔVbe than the good device.
Hurst tested several hundred devices and then subjected them to SAM analysis to detect the presence of die mount voids. He showed a clear correlation between the ΔV be measurements and the SAM analysis, and thus that thermal transient testing is a good proxy for die integrity as measured by SAM.
Poppe and Székely give an in-depth look at the power of similar techniques in Dynamic Temperature Measurements: Tools Providing a Look into Package and Mount Structures.
If money is no object there are yet other techniques for measuring temperature variations over an extremely small area. See Altet, Grauby and Volz Advanced Techniques for IC Surface Temperature Measurement.
[ back to top ]
This section contains brief descriptions of some of the measurements other than temperature that are needed in order to be able to create an accurate model of an application. As with the section on temperature measurement, in many cases we have indicated the issue and suggested appropriate reading material. You may not need detailed information immediately, but would benefit from becoming familiar with the issues, so please resist the temptation to skip by all the links!
The description of measuring methods for CTE, density and thermal conductivity given below has been summarised from Korb and Neubauer’s paper Thermophysical properties of metal matrix composites which is available at the MMC-Assess website, as Vol. 7 in their Metal Matrix Composites in Industry series (PDF file, 336KB). Whilst applied in their context to MMCs, the general principles are those used for all materials.
Much of the cooling in practical assemblies comes from moving air, and its velocity is a critical element in our calculations. Within a cabinet we can estimate flow indirectly from pressure drop, but there are several more direct methods:
The pitot tube is best suited to high rates of flow, and the laser methods are expensive to implement, so the mainstay of fluid measurement is the hot-wire anemometer in its various forms. Its principle is clearly described at the eFunda site at Hot-Wire Anemometers: Theory. As well as the classic design, which is an fragile hot wire, recent more robust designs use the cooling of a small heated probe surface.
As with many other types of measurement, we have to ask what effect making the measurement will have on the value of the parameter being measured. Also, for all types of anemometer, calibration is needed to ensure that the sensor output matches the velocity from a known source. Calibration is particularly difficult at low flow rates or in the presence of temperature gradients.
We recommend that you read the excellent review of flow-measurement techniques in the paper by Kaveh Azar Measuring fluid velocity in electronic enclosures.
CTE is usually measured using a dilatometer, where the sample is heated, applying a small force to a push rod, whose movement is detected by a displacement transducer, as shown in Figure 3. The method depends on reference materials, and requires some calibration, but can give results on CTE to within an error of 3–5%.
Density is determined by measuring the mass and volume of a specimen with a regular shape, usually cuboid. The alternative (pioneered by Archimedes!) is to measure the weight of a body both in air and suspended in a liquid in which the specimen will sink. The volume of the specimen can be determined by measuring the total volume, or by measuring the volume of liquid displaced. This method can be applied to specimens with irregular shapes or uneven surfaces, but the result may be distorted if the part is porous.
Generally we measure heat transfer only indirectly by measuring the temperature. But we can make estimates of the actual heat flux, as indicated in Ned Keltner’s paper Heat transfer measurements in electronics cooling applications. Note particularly his comments on sources of inaccuracy.
But how much heat is lost by convection? One mind-blowing but seemingly well-established way of determining this that uses the parallel between heat transfer and mass transfer is described in Roger Schmidt’s article Use of Naphthalene Sublimation Technique for Obtaining Accurate Heat Transfer Coefficients in Electronic Cooling Applications.
When it comes to measuring the parameters of thermal management materials, direct measurement of thermal resistance is possible, but presents practical difficulties, as described by Clemens Lasance in Problems with Thermal Interface Material Measurements: Suggestions for Improvement.
Remember how the performance of a material depends on how good a contact is made with the mating surfaces, so that greases often have surprisingly good thermal performance? One specific measurement problem lies in determining which element of the thermal resistance comes from the bulk material and which from the interfaces. An experimental procedure for doing this is explained by Bruce Guenin in Calculations for Thermal Interface Materials.
Specific heat is traditionally measured by differential scanning calorimetry (DSC). This is a technique for measuring how much energy is needed to establish a nearly zero temperature difference between a substance and an inert reference material, as the two specimens are subjected to identical temperature regimes in an environment heated or cooled at a controlled rate.
Two types of DSC systems are in common use (Figure 4). In power-compensation DSC, the temperatures of the sample and reference are controlled independently using separate, identical furnaces, and kept the same as the temperature rises by varying the power inputs. The energy required by the two furnaces at any temperature gives a measure of the heat capacity of the sample at that temperature relative to the reference.
In heat-flux DSC, sample and reference are connected by a low-resistance heat-flow path (a metal disc) and the assembly enclosed in a single furnace. Specific heat differences between sample and reference lead to a difference in their relative temperatures although the resulting heat flow is small because sample and reference are in good thermal contact.
DSC is also used to collect information on phase transitions in materials such as polymers, and on the ways in which metals cool. Visit http://www.pslc.ws/macrog/dsc.htm to see how the heat-flux method is used to examine thermal transitions in a polymer.
Thermal conductivity is more difficult to measure, but one method is indicated in Figure 5. The test specimen is sandwiched between two identical reference samples, and the stack placed between heating elements controlled at different temperatures. Using a guard heater around the stack to ensure a constant heat flux through the stack and no lateral heat losses, a steady-state temperature gradient is established along the stack and can be measured with thermocouples. The error for this method is quoted as approximately ±10%, with cylindrical specimens 25/50mm around 10–14mm thick.
Thermal conductivity can also be measured using a laser flash method. The laser fires a pulse at the front surface of the sample, as shown in Figure 6, and the infrared detector measures the temperature rise of the sample’s back surface. Such specimens are typically cylindrical, but only 1–3mm thick.
The software matches a theoretical curve to the experimental temperature-rise curve. The instrument can measure thermal diffusivity and specific heat simultaneously: the thermal diffusivity value is that associated with the nearest theoretical curve; for specific heat, the detector measures the actual temperature rise. The detector is calibrated with a reference sample of known specific heat. Measurements can be performed quickly, with an accuracy of about ±5%.
Even when software tools are used extensively, some form of measurement is always needed, because at some point the results presented by the simulation need to be validated against measurements in the equipment being modelled. Only by quantifying the actual errors can we ensure that the simulation has been based on the correct assumptions.
To check that you have a good understanding of the available methods of measuring temperature, airflow and other thermal-significant parameters, and of their limitations and sources of inaccuracy, we suggest you spend a little time thinking about how you would validate your thermal simulation of a typical set of board assemblies within an enclosure, such as a computer. What measurements are needed? What techniques would be appropriate? And how would you employ them?
Base your evaluation on your reading of selected resource materials as well as the discussion above.
There is no formal answer here, but these are thoughts to which we shall be returning in Unit 17.
[ back to top ]
Each of these lists is in the order in which the material is referenced in the Unit text. However, note that links to SAQ answers are not included!
[ back to top ]